RESEARCH DISCUSSION PAPER - Modelling Infl ation in Australia RDP 2010-03
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
2010-03
Reserve Bank of Australia
RESEARCH
DISCUSSION
PAPER
Modelling Inflation in
Australia
David Norman and
Anthony Richards
RDP 2010-03
Reserve Bank of Australia
Economic Research DepartmentMODELLING INFLATION IN AUSTRALIA
David Norman and Anthony Richards
Research Discussion Paper
2010-03
June 2010
Economic Analysis Department
Reserve Bank of Australia
We would like to thank Dougal Crawford for his contribution to the early stages of
this work, Adam Cagliarini, Adrian Pagan and Tim Robinson for their advice, and
David Jennings for his assistance with forecast comparisons. We also thank
Jacqui Dwyer, Alex Heath, Jarkko Jääskelä, Mariano Kulish, John Simon and
Anders Vredin for their comments. Responsibility for any remaining errors rests
with us. The views expressed in this paper are those of the authors and are not
necessarily those of the Reserve Bank of Australia.
Authors: normand or richardsa at domain rba.gov.au
Media Office: rbainfo@rba.gov.auAbstract
This paper estimates a range of single-equation models of inflation for Australia.
We find that traditional models, such as the expectations-augmented standard
Phillips curve or mark-up models, outperform the more micro-founded
New-Keynesian Phillips curve (NKPC) in explaining trimmed mean inflation, both
in terms of in-sample fit and significance of coefficients. This in large part reflects
the weak instruments problem in the estimation of the NKPC, and is partly
corrected by including a direct measure of inflation expectations, but we still find
that the unemployment rate or growth in marginal costs (unit labour cost and
import prices) provides a better fit than either the output gap or level of real
marginal costs. These traditional models also perform well in out-of-sample tests,
relative to alternative models and some common benchmarks, with the standard
Phillips curve clearly superior to these benchmarks on this test. As inflation has
become better anchored and hence less variable, the magnitude of the errors of the
single-equation models has declined, although the explanatory power (in terms of
R-squared) has fallen together with this greater stability. We also investigate the
empirical importance of some other variables that are commonly cited as
determinants of inflation, and find little evidence that either commodity prices or
the growth rate of money directly influence Australian underlying inflation.
JEL Classification Numbers: C53, E31
Keywords: inflation, modelling
iTable of Contents
1. Introduction 1
2. Modelling Structure and Theory 2
2.1 The ‘Standard’ or Expectations-augmented Phillips Curve 3
2.2 The New-Keynesian Phillips Curve 4
2.3 The Mark-up Model 5
3. Estimation 6
3.1 Data 6
3.2 Equation Specifications 10
4. Results 12
4.1 Main Results 12
4.2 Are There Benefits in Disaggregation? 17
4.3 Historical Fit of the Models 19
5. Investigating the Role of Other Variables or Restrictions 22
5.1 A Vertical Long-run Phillips Curve Restriction 22
5.2 Different Measures of Inflation Expectations 24
5.3 The Influence of Oil Prices 25
5.4 Monetary Models of Inflation 26
6. Changes in the Inflation Process over Time 27
7. Conclusions 30
Appendix A: Derivation of the Mark-up Model of Inflation 31
Appendix B: Additional Regression Results 34
References 35
iiMODELLING INFLATION IN AUSTRALIA
David Norman and Anthony Richards
1. Introduction
Central banks in most advanced economies now operate with some form of either
explicit or implicit inflation target, with the Reserve Bank of Australia (RBA)
operating under such a framework since 1993. A natural result of this change has
been an increased interest among central banks and academics in research about
the inflation process.
These efforts have resulted in a large number of different models for inflation,
which vary in terms of scale, and in terms of the variables assumed to drive
inflation (including the output gap, unemployment rate or unit labour costs). At the
RBA, a variety of approaches are used to model inflation, including single-
equation models (such as that in de Brouwer and Ericsson 1995), VAR models
(Gerard and Nimark 2008), factor models (Gillitzer and Kearns 2007; Gerard and
Nimark 2008) and DSGE models (Jääskelä and Nimark 2008). While there are
advantages to systems approaches to modelling inflation, single-equation models
can also be very useful due to their simplicity. This paper outlines the broad
structure of three types of single-equation models that are used as part of an array
of models at the RBA: a standard Phillips curve; a New-Keynesian Phillips curve;
and a mark-up model.
We find that the single-equation model approach has performed relatively well in
terms of both modelling and forecasting inflation over the past two decades, with
equal or better in-sample and out-of-sample performance than some standard
benchmarks and VAR models. The standard errors surrounding the model
predictions have also fallen since the introduction of inflation targeting. However,
it is notable that the explanatory power – in terms of adjusted R-squared – of these
equations has also declined over this time as inflation has become more stable. Of
the three models studied, the standard Phillips curve has performed best over this
time but the results from an expectations-augmented mark-up model are broadly
comparable. In contrast, the New-Keynesian Phillips curve model fits the data
relatively poorly, although most of this reflects issues with the use of generalised
method of moments (GMM) to instrument for expected inflation, and the model2 fits reasonably well when we use a direct measure of inflation expectations instead. Our results show that either the unemployment rate alone, or a combination of growth in unit labour costs and the output gap help to explain the deviation of inflation from measures of inflation expectations in Australia, once we have controlled for import price shocks. After controlling for the variables discussed above, we find little role for some other variables – notably commodity prices, excess money growth – that have sometimes been suggested as important determinants of inflation. A concern with these single-equation models, however, is that they do not impose many of the theoretical restrictions that would be standard in any structural model of inflation. For example, we find that imposing a restriction to ensure that the long-run Phillips curve is vertical results in a significant deterioration of fit for samples estimated since 1990. While there may be empirical issues or theoretical considerations that might explain this, such a result highlights the fact that reduced-form models such as those estimated in this paper are unlikely to perform well if inflation ever deviated persistently and substantially from target. Accordingly, it is appropriate to supplement the single-equation approach with results from more structural econometric models such as the Jääskelä-Nimark DSGE. The remainder of this paper is structured as follows. The various models are introduced in Section 2, the empirical estimation is discussed in Section 3 and the results are presented in Section 4. In Section 5 we assess the role of various additional variables and coefficient restrictions that are suggested by the existing literature. Section 6 assesses the stability of these models over time and Section 7 concludes. 2. Modelling Structure and Theory The literature identifies many ways to specify an inflation model. Setting aside the multi-equation and factor approaches discussed above, the structure of a single-equation inflation model can differ based on the variables assumed to drive inflation (most prominently, whether to use the output gap, the unemployment rate or marginal costs), how to model inflation expectations and whether to include foreign variables, among other considerations. Given this, we present variants of
3 three types of inflation equations in this paper – both ‘standard’ and New-Keynesian Phillips curve models, and a mark-up model. 2.1 The ‘Standard’ or Expectations-augmented Phillips Curve One category of inflation model is commonly referred to as the ‘standard’, or ‘expectations-augmented Phillips curve’ model. The original foundation for such models is the framework developed by Phillips (1958), which relates inflation to past and present values of some measure of resource utilisation such as the unemployment rate, so that reduced spare capacity contributes to higher inflation. Since the work of Milton Friedman, Edmund Phelps and others in the late 1960s and 1970s, it has been standard to augment this univariate relationship with inflation expectations, thereby avoiding the implication that there is a long-run trade-off between lower unemployment and higher inflation. Such models have been the workhorse of inflation modelling in a number of central banks over the past three decades. Empirical estimates of the standard Phillips curve can be found in a multitude of papers. One fairly standard specification relates inflation to several years of lagged inflation (as a proxy for inflation expectations), the current unemployment gap and measures of recent supply shocks (such as the relative price of food, energy and imports). Examples of this type of model include Gordon (2005) and Brayton, Roberts and Williams (1999). A slightly different approach can be seen in papers such as Gruen, Pagan and Thompson (1999), which include a direct estimate of inflation expectations derived from the bond market. Our specification follows this latter approach by directly including (bond market) inflation expectations. We also including the change in the unemployment rate to reflect ‘speed limit’ terms, and allow for only one supply shock – import price growth. However, our approach differs from Gruen et al (1999) in two respects. First, we do not allow for a time-varying non-accelerating inflation rate of unemployment (NAIRU) (although we assess the relevance of this simplification in Section 6). Second, we do not constrain our results to ensure a vertical long-run Phillips curve (although we consider the impact of this choice in Section 5.1). This results in the following standard Phillips curve model:
4
J K
t c Et 1 t s 1ur j urt j k mpt k t (1)
t i j 1 k 1
Where: π, ur and mp represents inflation, the unemployment rate and import prices,
respectively; Δ is the one-period change operator; and Et–1πt+s represents
expectations of inflation over the next s periods, formed in period t–1.
2.2 The New-Keynesian Phillips Curve
A second type of model that is widely used is the New-Keynesian Phillips curve
(NKPC; see Galí and Gertler 1999, among others). This builds on the
expectations-augmented Phillips curve by explicitly deriving the model from
microeconomic relationships between capacity utilisation, costs, prices and
nominal rigidities at the firm level. The resulting inflation equation then links the
deviation of inflation from its expected level to either the output gap or real
marginal cost.
The derivation of the closed economy version of the NKPC is now standard;
interested readers should consult Galí (2008) for details. In an open economy
setting, the derivation is similar except that it is appropriate to view consumer price
inflation as determined by a weighted average of domestic and imported inflation,
with the latter equal to the change in real import prices and the former determined
as per the standard closed economy NKPC, but with marginal cost proxied by both
real unit labour costs and real import prices. 1 The resulting open economy NKPC
can then be expressed as:
t Et t 1 ulct pt mpt pt mpt pt t (2)
where: ulc and mp represent nominal unit labour costs and import prices (world
prices, converted into domestic currency), respectively; p is the consumer price
level; and all variables are expressed as deviations from steady state. In empirical
work it is now also relatively standard to include a backward-looking inflation
term in the NKPC to produce a ‘hybrid’ NKPC, drawing on the work of Fuhrer and
Moore (1995).
1 The inclusion of both the level and change in real import prices allows for a general
specification, whereby imports can be either consumption or intermediate goods.5
An alternative specification is to use a measure of the output gap as a proxy for
real unit labour costs, given that these are proportional under certain conditions
(most notably that the capital stock is exogenously determined). However, it has
been a point of considerable controversy as to whether including the output gap
– defined as the deviation of actual GDP from its potential level – as the driving
variable for inflation provides a better representation than other proxies for real
marginal cost. 2 This may reflect the possibility that either the output gap is too
imperfectly measured to be a useful proxy for real unit labour costs, or that the
conditions under which they are proportional do not hold. We take an agnostic
view on this debate and include both the output gap and real unit labour costs in
our estimated equations. Theoretically this can be motivated by allowing the output
gap to influence inflation (over and above its influence on unit labour costs) via
either the cyclicality of non-wage costs (as in Leith and Malley 2007) or a
procyclical effect on the mark-up of price over marginal cost (that is, firms may
raise their mark-up over marginal cost when demand for their products is high).
As a result, our NKPC model can be specified as:
t Et t 1 1 t 1 rulct rmpt rmpt ygapt t (3)
where: rulc, rmp and ygap are real unit labour costs, real import prices and the
output gap, respectively; and all variables are expressed as deviations from their
sample means.
2.3 The Mark-up Model
Much of the previous Australian work on inflation modelling has tended to focus
on mark-up models of inflation rather than the Phillips curve approach (see, for
example, de Brouwer and Ericsson 1995, Heath, Roberts and Bulman 2004 and
Bårdsen, Hurn and McHugh 2007). In this model, inflation is determined by
current and lagged growth in unit labour costs and import prices, based on the
theory that firms set their prices as a mark-up on costs, as in the NKPC. However,
there has previously been little explicit allowance for forward-looking behaviour
2 Key players in this debate are Galí, Gertler and López-Salido (2005), who argue in favour of
the marginal cost-based NKPC, and Rudd and Whelan (2005), who argue in favour of the
(flex-price) output gap-based NKPC.6
by firms in such models, and the presence of nominal rigidities has only been
included implicitly by allowing for lags of input costs.
Nonetheless, it is straightforward to show that the NKPC can be rearranged to
reflect an expectations-augmented mark-up model. Full details of this are available
in Appendix A, but in short, this involves rearranging the NKPC so that it includes
terms for the growth in nominal marginal costs. We also allow for the mark-up to
be time-varying and positively related to the output gap. Given no change in the
inflation target, this results in the following specification for inflation:
J K
t Et 1 t s j ulct j k mpt k ygapt t 1 t (4)
j 0 k 1
where the notation is as per Equations (1) and (3). 3
3. Estimation
The estimation of these models raises a number of challenges both in terms of the
dependent variable, the functional form (for example, the lag structures used in the
regressions and the modelling of expectations) and the choice of particular
variables used in the regressions.
3.1 Data
Empirical papers on inflation in different countries typically use headline measures
of inflation (based on either the consumer price index or the household
consumption price index). Instead, we use an underlying measure: quarterly
trimmed mean inflation, where 15 per cent (by weight) of the CPI is trimmed from
the upper and lower ends of the distribution of seasonally adjusted price changes in
each quarter. 4 This measure is preferable to headline CPI since it provides a
3 Bårdsen et al (2007) argue that prices and real wages should be modelled as a system, to
reflect the fact that wage growth is dependent on prices and hence endogenous in
Equation (4). However, this should only be a serious econometric problem if the
contemporaneous correlation is significant, which does not appear to be the case.
4 The measure of trimmed mean inflation is based on data which are adjusted for the impact of
the tax changes of 1999–2000. Mortgage interest charges are omitted from the calculation of
the trimmed mean prior to September 1998, and the deposit & loan facilities item is excluded
from 2005.7
measure of underlying inflation that abstracts from much of the noise in CPI
inflation (Figure 1; see also Brischetto and Richards 2006).
Figure 1: Australian Inflation(a)
% Quarterly %
3 3
CPI
2 2
1 1
0 0
Trimmed mean
% %
Year-ended
12 12
9 9
6 6
3 3
0 0
1984 1989 1994 1999 2004 2009
Note: (a) Excluding mortgage interest charges prior to September quarter 1998 and deposit & loan facilities
prices since September 2005, and adjusted for the effects of tax changes of 1999–2000
Sources: ABS; RBA
Our measure of inflation expectations is derived from pricing in the bond market,
and is calculated as the difference between standard and indexed bond yields at a
maturity of approximately 10 years. (Alternative proxies are considered in
Section 5.2, and some candidate measures are shown in Figure 2.) There are
problems with this measure, including the fact that it includes any shifts in
liquidity premia, its long duration (ideally we would use a one- or two-year-ahead
expectation) and the fact that it is not directly observed from financial prices prior
to 1993 (see Gruen, Robinson and Stone 2002). 5 However, there are no obviously
superior measures and it proves to have some empirical explanatory power.
We estimate nominal unit labour costs as total labour costs divided by real GDP;
we use this in preference to the measure published by the ABS which is affected by
volatility in the estimate of self-employed income. For real unit labour costs, we
5 Prior to 1993, inflation expectations are estimated as the nominal bond yield less an assumed
real bond yield based on data for other OECD countries (see Gruen et al 2002).8 deviate from the practice in some of the US literature of using labour’s share of total income, given the substantial influence that commodity prices have on the GDP deflator (and hence labour’s share) in Australia. Instead, we calculate real unit labour costs by deflating nominal unit labour costs by trimmed mean inflation. Data on import prices are in local currency and are sourced from the ABS balance of payments release, but are adjusted in three ways. First, we remove ‘lumpy’ imported goods (large one-off imports such as civil aircraft) and oil (due to its volatility) from the calculation to produce an underlying measure of import prices that is more likely to be representative of persistent economy-wide import prices. Second, as in some earlier papers, we remove automated data processing equipment imports due to concerns about the treatment of quality changes when measuring the prices of such imports. Third, we adjust the across-the-docks import price data to include the effect of changes in tariff rates. Estimates of the output gap are based on a production function approach, in line with the practice favoured by the Organisation for Economic Co-operation and Development (OECD); although, broadly similar results are obtained when we use the capacity utilisation measure from the National Australia Bank business survey or a Hodrick-Prescott filtered output gap. The production approach calculates potential output growth as the weighted average of smoothed growth in the net capital stock and labour inputs (where the weight on each is the capital and labour share of total factor income, respectively) plus smoothed growth in multifactor productivity. This potential growth rate is then converted to a level of potential GDP by assuming that potential GDP was equal to actual GDP on average between 1998 and 2001 (in other words, that the output gap averaged zero over this time); the latter assumption has no effect on the regression results since the output gap is demeaned in each case.
9
Figure 2: Time Profile of the Determinants of Inflation
Sources: ABS; Melbourne Institute; RBA
Data for all these variables are sourced from the ABS, with the exception of
trimmed mean inflation, which is from the RBA, and inflation expectations, which
are from the RBA since 1993 and Gruen et al (2002) prior to that. 6 All data are
included in log or log-differenced form, except the unemployment rate which is in
levels.
The main estimation period is from the March quarter 1990 to the December
quarter 2009. The rationale for excluding data prior to 1990 is that a priori there is
the possibility of a structural break around the disinflation of the early 1990s and
the introduction of inflation targeting in 1993. We find evidence in support of this
(see Section 6), with rolling Chow tests clearly signalling a break in our inflation
equation around March 1989 (particularly in the standard Phillips curve model).
The decision to include data from 1990 rather than 1993 reflected a desire to use a
slightly longer sample, including the recession at the start of the 1990s, which may
improve our ability to accurately estimate coefficients.
6 Import prices are from the balance of payments release (ABS Cat No 5302.0); total labour
costs and GDP are from the quarterly national accounts release (ABS Cat No 5206.0); and
estimates of the capital stock and MFP are interpolated from the annual estimates published in
ABS Cat No 5204.0.10
3.2 Equation Specifications
The main choices regarding the specification are the appropriate econometric
technique to use when estimating each equation, and whether to impose any
constraints on the coefficient estimates. There are also questions about the
appropriate specification of the lag structure for the right-hand-side variables. In
particular, given that several of the variables are likely to feed through into
inflation with relatively long lags, it is necessary to consider lag structures which
preserve degrees of freedom. 7
The approach we use is to estimate each model using various permutations,
including in the way that lag structures are modelled. However, for simplicity, we
present results from only one econometric specification for each type of model.
Having decided on a method to preserve degrees of freedom, the exact lag length
and whether to allow speed limit effects or not were chosen by minimising the
Akaike information criteria (AIC). 8
Using this approach leads us to adopt the following standard Phillips curve
specification:
12
t c Et 1 t s 1 4ur mp
t 3 k t k t (5)
urt 2 k 1
where the coefficients on import prices (γk) are constrained to follow a quadratic
polynomial function in order to preserve degrees of freedom. The inverse of the
unemployment rate is included on account of it both providing the best fit and
being theoretically consistent with an increasing effect on inflation from a given
change in unemployment as it declines. Given the fact that the unemployment rate
is reasonably persistent (both in level and four-quarter-change form), there is a
7 Approaches used in the inflation modelling literature include: (i) estimation of an
unconstrained lag structure, with insignificant lags omitted based an on information criterion
or some other testing strategy; (ii) inclusion of long lags but constraints on some of the
coefficients to be the same, an approach followed by Robert Gordon in a number of papers,
among others; and (iii) the assumption that lag coefficients follow a polynomial function (or
Almon lags).
8 The AIC was used in preference to the Schwarz Bayesian information criterion – which more
heavily penalises additional parameters – since we are more interested in finding the
best-fitting model than in parsimony.11
range of feasible lag selections which yield broadly similar results: here we include
the unemployment rate with a two-quarter lag as suggested by the data (which may
reflect the presence of nominal rigidities in both wage- and price-setting) and a
lagged speed limit term as well. We include the prior period’s expectation of future
inflation to prevent estimation problems associated with endogeneity. Up to three
years of lags of import price growth are selected by the AIC. 9 Lagged inflation is
not found to be significant in this model and so is excluded from the estimation.
We estimate the NKPC using the GMM approach, as in Galí and Gertler (1999).10
As instruments, we use four lags of inflation and two lags of real unit labour costs.
We include only the current observation of the driving variables, so that our final
specification is as in Equation (3). We also estimate the NKPC using direct
measures of inflation expectations and ordinary least squares (OLS), as per Henzel
and Wollmershäuser (2008), in order to avoid having to instrument for inflation
expectations. For this specification, we include the prior period’s expectation of
future inflation and the first lag of all other variables to prevent endogeneity, and
we do not constrain the coefficients on inflation and inflation expectations as per
Equation (3), so that the specification is:
t Et 1 t s t 1 rulct 1 rmpt 1 rmpt 1 ygapt 1 t (6)
The main decision when estimating the mark-up model is which econometric
specification to use. An error correction specification has been popular in the past,
given the theoretical long-run link between the price level and marginal costs. But,
as discussed in Stone, Wheatley and Wilkinson (2005), it is difficult to determine
whether a cointegrating relationship exists between the price level, unit labour
costs and import prices without including another trending variable. Given this, our
preference is for distributed lag-type models. Within this class of models, we
investigate constraining the coefficients on input costs using either successive
four-quarter averages (as in Gordon 2005) or a polynomial distributed lag (PDL)
9 The appropriate number of lags of imports prices was chosen prior to imposing the
polynomial constraint, in order to ensure that the constraint itself did not influence the number
of lags.
10 Rudd and Whelan (2005) criticise the use of GMM on the grounds that any instrument used to
proxy future inflation is plausibly correlated with an omitted variable and could therefore bias
the estimated coefficients. However, this claim is disputed in Galí, Gertler and López-Salido
(2005) and it is still common to use the GMM approach.12
specification (as in Brayton et al 1999). We also estimate the coefficients without
any constraints. While the data provide mixed results on the validity of these
constraints, we are wary of over-fitting the data (which can occur with the
unconstrained model) and the constrained specification provides impulse responses
that are easier to interpret. Accordingly, our preferred specification of the mark-up
model is:
9 12
t c Et 1 t s j ulct j k mpt k ygapt 1 t (7)
j 0 k 1
where the coefficients λj and γk are constrained to follow a quadratic PDL function.
Lagged inflation is found to be insignificant in this specification and so is
excluded, and while the change in the output gap is significant at the 10 per cent
level, its coefficient is negative and so we exclude it from the equation.
4. Results
4.1 Main Results
The estimates of each model are shown in Table 1 (fitted values are shown in
Section 4.3). Overall, we find that the models fit the data quite well. They explain
around 60 per cent of the variation in quarterly underlying inflation and have root
mean squared errors (RMSE) that are generally less than 0.2 percentage points. In
terms of the in-sample performance of each model, we find that the standard
Phillips curve model fits best, followed by the mark-up model and then the two
NKPC models. This is evident from assessing each model’s AIC statistic, which is
clearly lowest for the standard Phillips curve (indicating superior fit) and highest
for the NKPC estimated with GMM.13
Table 1: Inflation Model Estimates
Standard PC NKPC OLS NKPC Mark-up model
Constant –0.009** 0.000 0.012***
Lagged inflation 0.163 0.148
Inflation expectations 0.403*** 0.837*** 0.361*** 0.309***
Output gap 0.020 0.166*** 0.178***
Unemployment rate 0.142***
Δ unemployment rate –0.003***
Real unit labour costs 0.039 0.125***
Δ unit labour costs 0.172*
Real import prices 0.006 0.027**
Δ import prices 0.079*** –0.011** –0.008 0.107***
Adjusted R2 0.648 0.585 0.618
AIC –9.91 –9.23 –9.75 –9.81
Log-likelihood 403.5 368.7 396.9 401.4
In-sample RMSE 0.156 0.311 0.171 0.160
Out-of-sample RMSE(a) 0.188 0.299 0.254 0.231
Durbin-Watson statistic 2.14 2.04 2.00
Notes: ***, ** and * represent significance at the 1, 5 and 10 per cent levels respectively. Where multiple lags
are included, coefficients shown are the sum of the lags. Coefficients on the constant, output gap, level
and change in unemployment rate, real unit labour costs and real import prices are multiplied by 4 to
represent annualised effects.
(a) Six quarters ahead, using real-time internal RBA forecasts of dependent variables and calculated over
the period since 2004.
The relatively poor performance of the NKPC estimated with GMM is also evident
when we look at the individual coefficients. We do not find significant coefficients
on any variable in the NKPC, except those relating to inflation and the change in
import prices (though the latter is incorrectly signed). This finding is robust to
different specifications of the driving variables, including estimating the model
with the output gap and real unit labour costs separately, and deflating real unit
labour costs by other price measures (including the gross national expenditure
deflator). This result should perhaps not be surprising, however, given that other
authors estimate insignificant coefficients on the driving variables in NKPCs based
on Australian data (for example, Kuttner and Robinson 2010).14 The results for the NKPC using OLS and direct measures of inflation expectations are more favourable. In particular, the coefficients on the driving variables are all now significant and sizeable (except for that on the change in import prices), and the overall fit of the model is markedly better than that estimated with GMM. However, the model does not fit the data as well as the standard Phillips curve and mark-up models. In contrast to the GMM-based NKPC, the coefficients on both the standard Phillips curve and the mark-up model are all correctly signed and significant. For unemployment, the coefficients on the (inverse) unemployment rate and the change in the unemployment rate imply that (other things equal) a 1 percentage point decline in unemployment (from 5 per cent) would raise the (quarterly annualised) inflation rate by around ¾ of a percentage point for four quarters (beginning two quarters after the initial change in unemployment), before the speed limit effect fades and the inflation rate subsequently settles back around ½ of a percentage point above its initial rate. The estimated coefficients on inflation expectations in the two models are broadly similar (at 0.3–0.4), as are the elasticities of inflation with respect to import prices (at 0.1). The estimated elasticity of inflation with respect to unit labour costs in the mark-up model is around two-thirds larger than this. The coefficient on the output gap implies that a gap of 1 per cent has a direct effect of 0.2 percentage points on annualised inflation, though this does not reflect the total effect since the output gap term here captures only its effects on inflation through non-wage costs and the mark-up, and not its effects on inflation via wage costs and expectations. The effects on inflation of movements in unit labour costs and import prices in these models are shown graphically in Figure 3. With regard to unit labour costs, the mark-up model implies that, other things equal, a 2½ per cent rise in unit labour costs (roughly corresponding to the standard deviation in annualised terms) would raise the (quarterly annualised) inflation rate by around 0.4 percentage points at its peak, which occurs around five to six quarters after the increase in unit labour costs. (Assuming no effect on inflation expectations or other variables, the effect on inflation is estimated to wash out after two years.) With regard to import prices, both the mark-up and standard Phillips curve models imply a very protracted pass-through to inflation. The mark-up model implies that the inflation rate would increase by a little more than ⅓ percentage point over each of the second and third years following a 10 per cent rise in import prices before washing
15
out after almost four years. 11 Likewise, the standard Phillips curve model implies a
three-year effect on inflation from higher import prices, with the estimated rise in
the inflation rate being around ⅓ percentage point over the second year and
slightly less in the third year. The protracted nature of these import-price effects
may partly reflect inventory behaviour (firms typically clear older stock acquired at
the original exchange rate before repricing). They may also reflect a deliberate
response to volatility in the exchange rate (with some firms absorbing changes in
import prices into their margins until they perceive that any exchange rate shock is
permanent). However, it is not clear that such explanations can explain the sizeable
effect on inflation that is ongoing three years after the shock. Nonetheless, this
result is robust across a range of model specifications and time periods.
Figure 3: Estimated Response of Inflation to Regressors(a)
Annualised quarterly inflation rate
Note: (a) Assuming (permanent) 2½ per cent change in level of unit labour costs and 10 per cent change in
import prices
11 Dwyer and Leong (2001) suggest that just over half of any shock to the exchange rate is
estimated to flow into across-the-docks import prices within the quarter, with the remainder
flowing through in the following quarter.16 In order to assess the performance of these models, it is useful to compare their out-of-sample forecasts with those of some standard benchmarks. In particular, we compare these models to four alternatives: a constant annualised rate of 2½ per cent (the mid-point of the inflation target); the previous quarter’s year-ended underlying inflation rate; a simple vector autoregression (VAR) model that uses the same variables that are included in the mark-up model; 12 and the factor-augmented VAR (FAVAR) model developed by Gerard and Nimark (2008). Real-time estimates of the regressors in the single-equation models are taken from internal RBA forecasts; as such, this exercise is not a ‘pure’ test of the forecasting ability of single-equation models, since forecasts of the independent variables are derived in a different manner from that used by the VAR or FAVAR models. With this caveat in mind, this exercise suggests that the single-equation models generally perform better than these benchmarks. Table 2 presents the ratio of the root mean squared forecast error (RMSFE) of each model to that of a constant at various horizons, calculated between 2004 and 2009. All models perform somewhat better than the constant forecast, as reflected in ratios of less than one. 13 The standard Phillips curve model has the lowest RMSFE at all horizons, outperforming the benchmark models by at least 10 per cent, and considerably more at longer horizons. The performance of the mark-up model is very similar to each of the benchmarks at shorter horizons, but modestly superior at the eight- quarter-ahead horizon. The NKPC model estimated with OLS produces better forecasts over the long horizon than the benchmarks, but tends to produce inferior results over a shorter horizon. 12 To conserve degrees of freedom, only one lag of each endogenous variable (trimmed mean inflation, inflation expectations, the output gap and quarterly growth in unit labour costs) is included, but the model is augmented with various lags of year-ended unit labour costs and import price growth to capture the same lag length as in the mark-up model. 13 The exception is the GMM-estimated NKPC (not shown), whose forecasts are found to be no better at all horizons.
17
Table 2: Relative Forecasting Performance
Ratio of root mean squared forecast error to that from constant forecast(a)
Single-equation models Benchmark models
OLS Standard Mark-up Lagged Simple Factor-
NKPC PC model inflation VAR augmented
VAR
1 qtr ahead 0.834 0.708 0.856 0.790 0.803 0.764
4 qtrs ahead 0.845 0.640 0.801 0.742 0.762 0.843
8 qtrs ahead 0.810 0.590 0.818 0.872 0.863 1.009
Notes: (a) For the period 2004–2009. RMSFE for 1-quarter-ahead horizon calculated from quarterly inflation
rate, while 4- and 8-quarter-ahead RMSFEs are based on the year-ended inflation rate.
4.2 Are There Benefits in Disaggregation?
An alternative approach to that used so far is to disaggregate inflation into some
combination of the 90 items in the CPI, and estimate models for these
disaggregated components. This allows the coefficients on input costs or
expectations to vary across different types of products, although it may come at the
cost of additional noise in the dependent variable. We split the sample of 90 items
into two groups. The first includes items with prices predominantly determined by
external factors (most importantly, the exchange rate). This largely consists of the
manufactured items within the CPI, along with overseas travel; we loosely refer to
this as the ‘manufactures’ group. The second group includes items with prices
largely determined by domestic considerations, and we refer to this as the ‘non-
manufactures’ group. 14 Based on this classification, we calculate trimmed mean
inflation for each series and use this as the dependent variable.
In each case we estimated both a standard Phillips curve and a mark-up model of
inflation. We present the models which fit best, with the mark-up specification
14 The main manufactured items excluded from the manufactures group are alcohol & tobacco,
pharmaceuticals, and books & newspapers (whose prices appear to be largely determined
domestically). Automotive fuel was excluded from both groups, as its weight dwarfs all other
items in the manufactures classification where it would naturally fit. (This introduces an
inconsistency with the earlier results, but it is not a material one.) Deposit & loan facilities
were also excluded from the estimation, consistent with our approach at the aggregate level.
The resulting grouping has some similarity to the tradable/non-tradable split published by
the ABS, with the main difference being that a range of food items that are tradable are
included in our non-manufactures series since their prices appear to be largely influenced by
domestic factors.18
clearly superior in the case of manufactured items and the standard Phillips curve
marginally superior for non-manufactured items. The model for manufactures
inflation is:
12
t
M
c M
Et 1 t s ulct 1 kM mpt k t
M M 4
(8)
k 1
where: the superscript M refers to the group of manufactured items; input costs and
inflation expectations are economy-wide measures (due to data limitations); and
the coefficients on import prices are constrained to follow a quadratic PDL
specification. 15 For inflation of the non-manufactures group (NM), the model is:
1 12
t
NM
c NM
NM
Et 1 t s NM
urt 3 kNM mpt k t (9)
NM 4
urt 2 k 1
where the coefficients on import prices are also constrained using a quadratic PDL
specification.
The results of this are shown in Table 3. Not surprisingly, import prices are an
important explanator of manufactures prices, while domestic determinants
(inflation expectations and labour market conditions) are more important for
non-manufactures. The fitted values can be aggregated based on the relative share
of manufactured and non-manufactured items in the CPI. This combination of the
disaggregated models produces broadly comparable results to those from the
aggregated model, in terms of in-sample fit and out-of-sample fit. In addition, the
implied sensitivity of overall inflation to movements in input costs is also similar,
especially for import prices (at just over 0.1). These results suggest that the gains
from the disaggregate approach come primarily from its usefulness in illustrating
some underlying influences, with its econometric performance providing no strong
reason to prefer it.
15 The use of the year-ended change specification for unit labour costs is equivalent to the
inclusion of four quarterly lags that are constrained to be equal.19
Table 3: Disaggregated Inflation Model Estimates
Manufactured Non-manufactured Aggregate
items items results(a)
Constant –0.011** –0.003
Inflation expectations 0.358*** 0.425***
Unemployment rate (inverse) 0.140***
Δ unemployment rate –0.003***
Δ unit labour costs 0.117
Δ import prices 0.309*** 0.032
2
Adjusted R 0.395 0.684
In-sample RMSE 0.380 0.152 0.159
Out-of-sample RMSE(b) 0.477 0.209 0.246
Durbin-Watson statistic 2.42 1.75
Notes: ***, ** and * represent significance at the 1, 5 and 10 per cent levels respectively. Where multiple lags
included, coefficients shown are sum of the lags. Coefficients on constant and the level and change in the
unemployment rate are multiplied by 4 to represent annualised effects.
(a) RMSE is calculated as a weighted average of manufactures and non-manufactures regressions.
(b) Six quarters ahead, using real-time internal RBA forecasts of dependent variables and calculated over
the period since 2004.
4.3 Historical Fit of the Models
Figure 4 shows the year-ended fitted values of the NKPC (estimated with both
GMM and OLS), the standard Phillips curve, the mark-up model and the weighted
average of the models for manufactures and non-manufactures (referred to
hereafter as the ‘disaggregate’ model). To account for the lead of inflation in the
GMM-based NKPC, the fitted values from that model are estimated using the
closed form implied by the model, as in Galí and Gertler (1999) and
Sbordone (2002). 16
16 More specifically, we first estimate two recursive VAR models and a recursive autoregressive
model to forecast unit labour costs, import prices and the output gap respectively, in a similar
manner to Sbordone (2002). The VAR for unit labour costs (import prices) includes two lags
each of the growth rate of nominal unit labour costs (import prices) and the level of the
mark-up of consumer prices on unit labour cost (mark-up of consumer prices on import
prices). Each model is estimated from 1982:1 to period t, and is used to forecast the driving
variables 12 quarters ahead starting in t+1. The discounted sum of these forecasts is then
included in the closed form solution, as per Galí and Gertler (1999), with equations for the
characteristic roots found in Galí et al (2005).20
Figure 4: Fitted Values of Trimmed Mean Inflation Models
Year-ended
Note: (a) Estimated using closed form, corresponding to Galí and Gertler (1999) graph of ‘fundamental
inflation’
This evidence reinforces that obtained from the regression results that the standard
Phillips curve and mark-up models provide a better description of inflation than
does the NKPC. The standard Phillips curve and mark-up models track the cycle in
inflation more closely, and better capture the pick-up in inflation in 1995, 2001 and
2007–2008. Likewise, the disaggregate model provides a relatively good
explanation of aggregate inflation over much of the sample, with the fitted values
from it resembling those from the standard Phillips curve model at most points in
the sample. The relatively good performance of the disaggregate model reflects the
performance of both its components (Figure 5). The manufacture and
non-manufactures models both fit the data well, although the non-manufactures
model suggests an earlier and smoother pick-up in domestic inflation over
2005–2008 than actually occurred, and the manufactures model does not explain
the full extent of imported inflation that occurred in 2001.21
Figure 5: Manufactures and Non-manufactures Trimmed Mean Inflation
Year-ended
In contrast, the GMM-based NKPC model produces large and persistent errors.
This may in part reflect the simplicity of our forecasting models for the driving
variables, which do not capture much of the variation in these series, although it
likely also reflects the significant and persistent fluctuations in real import prices
over the sample, which have a longer-lasting effect on the NKPC than other
models. Using OLS to estimate the NKPC partially resolves these issues, with the
model having a closer fit. Nonetheless, the OLS version of the NKPC still displays
much larger errors than the other models.
All models correctly suggest a substantial acceleration in inflation between 2005
and 2008. Estimates from the standard Phillips curve are closest to the true
outcome, at around 4 per cent by late 2008, while those from the other models are
moderately lower than actually occurred, at around 3½–3¾ per cent. This
estimated pick-up is consistent with the trend decline in the unemployment rate
over that time and associated strength in unit labour costs growth, along with a
sizeable output gap and fading disinflation from import prices. All models also
explain the disinflation that occurred through 2009.22 5. Investigating the Role of Other Variables or Restrictions The literature on inflation models contains a much larger set of candidate variables than those used above. Furthermore, there is considerable debate about how to best proxy inflation expectations, and many other papers impose the restriction that there is no long-run trade-off between inflation and the unemployment rate or output gap, which we have not done. Given this, it is useful to check what effect such variations would have on our results. We pursue four issues: the effect of imposing a vertical long-run Phillips curve restriction on our models; whether there are better proxies for inflation expectations; what influence commodity prices have on inflation; and the possible role of money growth. 17 In each case, we adapt the standard Phillips curve, the OLS NKPC and the mark-up model with these restrictions or additional variables. The details of each regression can be found in Appendix B. 5.1 A Vertical Long-run Phillips Curve Restriction The concept of a vertical long-run Phillips curve – that is, that there is no trade-off in the long run between unemployment and inflation – is a cornerstone of most inflation models and has a long history in the literature (beginning with the rational expectations revolution in the 1970s). Indeed, this idea underlies the pursuit of price stability by central banks in most industrial countries. However, neither our standard Phillips curve, the mark-up model nor the OLS NKPC incorporate this standard feature. This reflects the fact that the sum of the coefficients on the right-hand-side nominal variables (inflation expectations, and growth in unit labour costs and import prices) in each of these models is significantly less than unity. This result survives when we simplify the models to represent inflation as being driven solely by inflation expectations and the output gap or unemployment rate; in that case, the coefficient on inflation expectations is significantly different from 1 for samples beginning in or after 1987. The rejection of this restriction is, however, not unique to our work; for example, both Anderson and Wascher (2000) and Williams (2006) have found such a result using US data. 17 We also looked at a possible role for the world output gap, but found no support for this (results available from the authors on request).
23 There are several possible explanations for this result. One relates to econometric issues, including from the relatively small sample used in estimation, or the possibility of a bias to our coefficients from measurement errors in our regressors. For example, our measure of expectations of inflation over the next 10 years is likely to be imprecisely measured (as are other measures of expectations) and only imperfectly correlated with the expectation that influences price-setting (which is likely to be for a one- or two-year horizon). However, it is hard to assess how important these effects are, and the long-run restriction is also obtained if we use a measure of shorter-term (one- or two-year) inflation expectations derived from the term structure of bond yields. Another possible explanation is suggested by Akerlof, Dickens and Perry (2000), namely that it may be ‘near rational’ for agents to devote limited attention to inflation when it is at low levels, resulting in the long-run Phillips curve appearing to be non-vertical within a range of low inflation outcomes. Regardless of the reasons, it is clear that the models estimated above are not ‘structural’, in that they would not be robust to examining episodes of high inflation. One response to such concerns would be to impose the property of long-run neutrality on our models. This can be done by respecifying the mark-up model with the coefficients on the nominal terms constrained to sum to unity, which is equivalent to deflating all nominal variables (including the dependent variable) by inflation expectations. However, this results in a deterioration of fit, with the adjusted R-squared falling from 0.62 to 0.45, and the introduction of mild serial correlation. This serial correlation arises because the coefficients on the input cost terms increases significantly, from 0.17 to 0.48 for unit labour costs and 0.11 to 0.27 for import prices, causing the model to consistently under-predict inflation following currency appreciations and over-predict inflation following currency depreciations. Nonetheless, this model might be more robust in the event that inflation expectations become dislodged and so may be a useful addition to any suite of models.
24 5.2 Different Measures of Inflation Expectations This discussion highlights some of the potential difficulties with the inflation expectations measure used in our models. While it is commonly accepted that inflation expectations are a critical influence on inflation outcomes, there is no consensus on how to model such expectations appropriately. The traditional approach has been to proxy inflation expectations with lagged inflation (as in Brayton et al 1999, for example), while the New-Keynesian approach of instrumenting for expected inflation using lags of inflation (as well as the output gap and unit labour costs) is essentially equivalent. Both of these, however, have well-understood limitations. An alternative approach is to use survey or financial market measures of inflation expectations (for example, Gruen et al 1999 and Henzel and Wollmershäuser 2008). We test the sensitivity of our results to these different approaches. The results imply that, for both the mark-up and standard Phillips curve models of inflation, the bond market measure provides the best in-sample fit. The results of Section 4 have shown that using instruments for inflation expectations is inferior to including the bond market measures of inflation expectations. The bond market measure also provides a better fit than lagged year-ended inflation; the explanatory power falls when we include lagged inflation instead (Table B1) and its coefficient is insignificant when included alongside the bond market measure (result not shown). The bond market measure also provides a superior fit to either the Melbourne Institute’s survey measure of consumer expectations (Table B1) 18 or measures of shorter-term inflation expectations implied by financial markets (result not shown). All of these results are invariant to whether the sample begins in 1990 or 1993. 18 One difficulty with using the Melbourne Institute’s measure is that it contains a structural break in 1993, prior to which negative responses were excluded from the calculation. This results in the current series being a full percentage point above the old series in 1993. To ensure consistency back to 1990, we assume that this gap remains constant from 1990 to 1993, and adjust down the old measure accordingly.
25 5.3 The Influence of Oil Prices Many papers modelling inflation include a measure of oil or other commodity prices in their equations (see, for example, Gordon 2005). In most instances, such regressions use headline rather than underlying (or ‘core’) inflation, providing a strong rationale for the inclusion of oil prices in the equation. 19 To investigate whether there is a role for oil prices in explaining underlying inflation in Australia, we augment our aggregate regressions with lagged growth in (Australian dollar) crude oil prices (or the level of real oil prices in the NKPC). We test for up to six years of lags, although to preserve degrees of freedom we constrain the quarterly responses to be equal within each year. The results from this exercise present mixed evidence on the role of oil prices (Table B1). The mark-up model provides support for a statistically significant effect of movements in oil prices; also, the overall explanatory power rises modestly compared with the baseline and a likelihood ratio test is able to reject the hypothesis that oil prices have no effect on underlying inflation at the 10 per cent level. 20 However, there are a number of concerns with these results, including the very long lag between oil price changes and underlying inflation estimated by the model (three to four years), sensitivity to the lag structure, and the insignificance of oil prices in the Phillips curve models. Overall, these results provide only limited 19 The international evidence on whether to include oil prices in a regression for underlying inflation is limited. One comprehensive assessment of this question is provided by Hooker (2002), who finds that while oil prices were a significant explanator of core inflation in the United States prior to 1980, they play an insignificant role thereafter. In contrast, Cheung (2009) finds that persistent movements in oil prices have been a useful predictor of core inflation since 1990 across a range of countries. While Hooker argues that his results stem from a change in policy by the Federal Reserve since 1980 to not accommodate oil price shocks, the difference between his results and those of Cheung suggests that it might instead reflect the relative stability of oil prices between 1980 and 2000. In other words, the sharp rise in oil prices between 2003 and mid 2008 could explain Cheung’s finding of significant second-round effects. 20 The elasticity of inflation with respect to movements in crude oil prices in the mark-up model is estimated to be 0.018 (not including any second-round effects, for example on inflation expectations) which implies that the more than doubling of oil prices in the four years to late 2008 contributed almost ½ a percentage point to annual inflation. Rolling regression estimates suggest that this relationship is only identifiable during periods of large changes in oil prices, since the coefficients are insignificant in samples that exclude the past few years.
26 support for the presence of an effect on underlying inflation from oil price growth, independent of its effect via inflation expectations or the output gap. 21 5.4 Monetary Models of Inflation There is a long tradition of modelling inflation as a function of growth in the money supply, based on Milton Friedman and Anna Schwartz’s observation that ‘inflation is always and everywhere a monetary phenomenon’. While this is no longer considered to imply that central banks should target growth in the money supply, there is still some debate about whether money should be considered when modelling inflation. For example, the European Central Bank’s ‘two-pillar’ monetary policy framework might suggest such an approach, relating short- to medium-term inflation to real factors (such as output and unemployment gaps) and medium- to long-term inflation to growth in the money supply. However, many economists of the New Keynesian tradition have argued to the contrary, suggesting that money is ‘obsolete’ in a properly specified model that includes a monetary policy rule (see, for example, Woodford 2007). Given these differences, it is of interest to assess whether money growth has any explanatory power for Australian inflation. This could arise for a variety of reasons unrelated to whether it is appropriate to include money in a structural model of inflation; for example, trend money growth could provide a better (or complementary) proxy for inflation expectations than the bond market measure, or might control for measurement error in unit labour costs. We use an approach adopted by Gerlach (2004), who augments a NKPC model with a measure of trend money growth. Gerlach motivates this specification on the basis that the NKPC captures short- to medium-term movements in inflation, while trend money growth will capture the medium- to long-run movements in inflation. We implement a version of this approach, but capture the short- to medium-term dynamics of inflation with the standard Phillips curve model. Specifically, we supplement Equation (5) with a term for money growth and estimate the following specification: 21 We also tested for the inflationary effect of commodity prices more generally, using the CRB commodity price index (in Australian dollar terms). We do not find a significant coefficient on year-ended growth in commodity prices over most sample periods.
You can also read
